• Publicationsclose
• Research dataclose
• Research softwareclose
• 2018-2022close
• Open Access close
• Preprint close
• AE close
• arXiv.org e-Print Archive close

Person re-identification has achieved great progress with deep convolutional neural networks. However, most previous methods focus on learning individual appearance feature embedding, and it is hard for the models to handle difficult situations with different illumination, large pose variance and occlusion. In this work, we take a step further and consider employing context information for person search. For a probe-gallery pair, we first propose a contextual instance expansion module, which employs a relative attention module to search and filter useful context information in the scene. We also build a graph learning framework to effectively employ context pairs to update target similarity. These two modules are built on top of a joint detection and instance feature learning framework, which improves the discriminativeness of the learned features. The proposed framework achieves state-of-the-art performance on two widely used person search datasets. Comment: To appear in CVPR 2019

Scenario Analysis is a risk assessment tool that aims to evaluate the impact of a small number of distinct plausible future scenarios. In this paper, we provide an overview of important aspects of Scenario Analysis including when it is appropriate, the design of scenarios, uncertainty and encouraging creativity. Each of these issues is discussed in the context of climate, energy and legal scenarios.

We propose a data-driven approach for deep convolutional neural network compression that achieves high accuracy with high throughput and low memory requirements. Current network compression methods either find a low-rank factorization of the features that requires more memory, or select only a subset of features by pruning entire filter channels. We propose the Cascaded Projection (CaP) compression method that projects the output and input filter channels of successive layers to a unified low dimensional space based on a low-rank projection. We optimize the projection to minimize classification loss and the difference between the next layer's features in the compressed and uncompressed networks. To solve this non-convex optimization problem we propose a new optimization method of a proxy matrix using backpropagation and Stochastic Gradient Descent (SGD) with geometric constraints. Our cascaded projection approach leads to improvements in all critical areas of network compression: high accuracy, low memory consumption, low parameter count and high processing speed. The proposed CaP method demonstrates state-of-the-art results compressing VGG16 and ResNet networks with over 4x reduction in the number of computations and excellent performance in top-5 accuracy on the ImageNet dataset before and after fine-tuning.

In this paper, we present a novel data structure, called the Mixture Graph. This data structure allows us to compress, render, and query segmentation histograms. Such histograms arise when building a mipmap of a volume containing segmentation IDs. Each voxel in the histogram mipmap contains a convex combination (mixture) of segmentation IDs. Each mixture represents the distribution of IDs in the respective voxel's children. Our method factorizes these mixtures into a series of linear interpolations between exactly two segmentation IDs. The result is represented as a directed acyclic graph (DAG) whose nodes are topologically ordered. Pruning replicate nodes in the tree followed by compression allows us to store the resulting data structure efficiently. During rendering, transfer functions are propagated from sources (leafs) through the DAG to allow for efficient, pre-filtered rendering at interactive frame rates. Assembly of histogram contributions across the footprint of a given volume allows us to efficiently query partial histograms, achieving up to 178$\times$ speed-up over na$\mathrm{\"{i}}$ve parallelized range queries. Additionally, we apply the Mixture Graph to compute correctly pre-filtered volume lighting and to interactively explore segments based on shape, geometry, and orientation using multi-dimensional transfer functions. Comment: To appear in IEEE Transacations on Visualization and Computer Graphics (IEEE Vis 2020)

A characteristic feature of the 3d plaquette Ising model is its planar subsystem symmetry. The quantum version of this model has been shown to be related via a duality to the X-Cube model, which has been paradigmatic in the new and rapidly developing field of fractons. The relation between the 3d plaquette Ising and the X-Cube model is similar to that between the 2d quantum transverse spin Ising model and the Toric Code. Gauging the global symmetry in the case of the 2d Ising model and considering the gauge invariant sector of the high temperature phase leads to the Toric Code, whereas gauging the subsystem symmetry of the 3d quantum transverse spin plaquette Ising model leads to the X-Cube model. A non-standard dual formulation of the 3d plaquette Ising model which utilises three flavours of spins has recently been discussed in the context of dualising the fracton-free sector of the X-Cube model. In this paper we investigate the classical spin version of this non-standard dual Hamiltonian and discuss its properties in relation to the more familiar Ashkin-Teller-like dual and further related dual formulations involving both link and vertex spins and non-Ising spins. Reviews results in arXiv:1106.0325 and arXiv:1106.4664 in light of more recent simulations and fracton literature. Published in special issue of Entropy dedicated to the memory of Professor Ian Campbell

We propose a new scaling ansatz in the neutrino Dirac mass matrix to explain the low energy neutrino oscillations data, baryon number asymmetry and neutrinoless double beta decay. In this work, a full reconstruction of the neutrino Dirac mass matrix has been realized from the low energy neutrino oscillations data based on type-I seesaw mechanism. A concrete model based on $A_4$ flavor symmetry has been considered to generate such a neutrino Dirac mass matrix and imposes a relation between the two scaling factors. In this model, the right-handed Heavy Majorana neutrino masses are quasi-degenerate at TeV mass scales. Extensive numerical analysis studies have been carried out to constrain the parameter space of the model from the low energy neutrino oscillations data. It has been found that the parameter space of the Dirac mass matrix elements lies near or below the MeV region and the scaling factor $|\kappa_1|$ has to be less than 10. Furthermore, we have examined the possibility for simultaneous explanation of both neutrino oscillations data and the observed baryon number asymmetry in the Universe. Such an analysis gives further restrictions on the parameter space of the model, thereby explaining the correct neutrino data as well as the baryon number asymmetry via a resonant leptogenesis scenario. Finally, we show that the allowed space for the effective Majorana neutrino mass $m_{ee}$ is also constrained in order to account for the observed baryon asymmetry. Comment: 25 pages, 10 figues, revised version

ABSTRACT We report the detection of a bright fast radio burst, FRB 191108, with Apertif on the Westerbork Synthesis Radio Telescope. The interferometer allows us to localize the FRB to a narrow 5 arcsec × 7 arcmin ellipse by employing both multibeam information within the Apertif phased-array feed beam pattern, and across different tied-array beams. The resulting sightline passes close to Local Group galaxy M33, with an impact parameter of only 18 kpc with respect to the core. It also traverses the much larger circumgalactic medium (CGM) of M31, the Andromeda Galaxy. We find that the shared plasma of the Local Group galaxies could contribute ∼10 per cent of its dispersion measure of 588 pc cm−3. FRB 191108 has a Faraday rotation measure (RM) of +474 $\pm \, 3$ rad m−2, which is too large to be explained by either the Milky Way or the intergalactic medium. Based on the more moderate RMs of other extragalactic sources that traverse the halo of M33, we conclude that the dense magnetized plasma resides in the host galaxy. The FRB exhibits frequency structure on two scales, one that is consistent with quenched Galactic scintillation and broader spectral structure with Δν ≈ 40 MHz. If the latter is due to scattering in the shared M33/M31 CGM, our results constrain the Local Group plasma environment. We found no accompanying persistent radio sources in the Apertif imaging survey data.

Ultra-dense deployments in 5G, the next generation of cellular networks, are an alternative to provide ultra-high throughput by bringing the users closer to the base stations. On the other hand, 5G deployments must not incur a large increase in energy consumption in order to keep them cost-effective and most importantly to reduce the carbon footprint of cellular networks. We propose a reinforcement learning cell switching algorithm, to minimize the energy consumption in ultra-dense deployments without compromising the quality of service (QoS) experienced by the users. In this regard, the proposed algorithm can intelligently learn which small cells (SCs) to turn off at any given time based on the traffic load of the SCs and the macro cell. To validate the idea, we used the open call detail record (CDR) data set from the city of Milan, Italy, and tested our algorithm against typical operational benchmark solutions. With the obtained results, we demonstrate exactly when and how the proposed algorithm can provide energy savings, and moreover how this happens without reducing QoS of users. Most importantly, we show that our solution has a very similar performance to the exhaustive search, with the advantage of being scalable and less complex. 30 pages, 7 Figures, Submitted to IEEE Transactions on Green Communications and Networking (IEEETGCN)

In this remark, we identify the cause of the loss of accuracy in the computation of the Faddeyeva function, w ( z ), near the real axis when using Algorithm 680. We provide a simple correction to this problem that allows us to restore this code as one of the important reference routines for accuracy comparisons.

We present the second data release (DR2) of the X-Shooter Spectral Library (XSL), which contains all the spectra obtained over the six semesters of that program. This release supersedes our first data release from Chen et al. 2014, with a larger number of spectra (813 observations of 666 stars) and with a more extended wavelength coverage as the data from the near-infrared arm of the X-Shooter spectrograph are now included. The DR2 spectra then consist of three segments that were observed simultaneously and, if combined, cover the range between $\sim$300 nm and $\sim$2.45 $\mu$m at a spectral resolving power close to $R=10\,000$. The spectra were corrected for instrument transmission and telluric absorption, and they were also corrected for wavelength-dependent flux-losses in 85% of the cases. On average, synthesized broad-band colors agree with those of the MILES library and of the combined IRTF and Extended IRTF libraries to within $\sim\!1$%. The scatter in these comparisons indicates typical errors on individual colors in the XSL of 2$-$4 %. The comparison with 2MASS point source photometry shows systematics of up to 5% in some colors, which we attribute mostly to zero-point or transmission curve errors and a scatter that is consistent with the above uncertainty estimates. The final spectra were corrected for radial velocity and are provided in the rest-frame (with wavelengths in air). The spectra cover a large range of spectral types and chemical compositions (with an emphasis on the red giant branch), which makes this library an asset when creating stellar population synthesis models or for the validation of near-ultraviolet to near-infrared theoretical stellar spectra across the Hertzsprung-Russell diagram. Comment: Main paper: 16 pages, Full list of stars from pages 21 to 34, Typical XSL spectra on Figure 10 page 10